Date:
Thursday , March 09, 2006
Author:
Brent Justice
Editor:
Kyle Bennett
Google +1

NVIDIA GeForce 7900 Series Preview

NVIDIA’s new GeForce 7 series GPUs are poised to provide you with an Extreme High-Definition gaming experience. We’ll talk about what is new, what isn’t, and show you what it all boils down to in terms of the real gameplay experience provided.

continued...

The Same

NVIDIA is announcing the G71 which will be known as the GeForce 7900 GTX and GeForce 7900 GT. They are also announcing the G73 which will be known as the GeForce 7600 GT.

Take everything you know about the G70 series and apply it to these new GPUs. The architecture is exactly the same.

Article Image

That means there are 8 vertex units in the G70, same for the G71, the G70 can do 24 pixels per clock, so can the G71, the G70 has a total of 48 ALUs available to the pixel rendering part of the GPU, so does the G71, there are 16 ROPs in the G70, same with the G71. Take all of that and cut it in half and you have the G73.

Here is a table that sums it all up.

Article Image

Along with the same architecture, the features are the same, Shader Model 3.0 support, FP16 HDR support, NO AA with FP16 HDR. You get the same Antialiasing and Anisotropic support, Transparency Multi and Supersampling at 2X, 4X and an 8xS mode, and Anisotropic up to 16X with the same angle-dependent AF.

We asked NVIDIA if this angle-dependency is a hardware limitation or a software one, they responded that this is a hardware issue. I don’t think we’ll be seeing any driver that will help in this area. We did learn that future generation video cards may not have this limitation, though nothing concrete on any details. Obviously NVIDIA don’t comment on unannounced products, but we think it likely that future cards may have better filtering, at least we hope.

The New

Now we can look at what does separate the G71 and G73 from the rest of the crowd.

Article Image Article Image Article Image

As we mentioned in the intro the biggest change is the move to 90 nanometers. This move has allowed NVIDIA to do some re-pipelining of the GPU that actually results in LESS transistors. If you aren’t familiar with what “re-pipelining” means think of it like this; there are certain ways to lay out your pipelines and stages in a GPU or any processor. Given a certain die size and manufacturing process you have to lay them out accordingly, and in your first generation of figuring out how many stages and the like to build your pipelines you may include redundant or un-optimized stages in a pipeline. As you learn more about your GPU you may figure out a way to lay down the circuitry in a more efficient manner reducing redundant processes, which means reducing transistors.

What it all boils down to is that with this new manufacturing process and further tweaking of their architecture NVIDIA have managed to eliminate transistors that aren’t needed and reduced stages in the pipelines. What may have taken for example three stages, can now maybe be done in one. This simply means good things for me and you in terms of power, heat and frequency.

Where this reduction will further come into play is in laptops. Laptops need the most power efficient designs in order to operate with less exotic cooling solutions and to save battery life.

NVIDIA also made some very minor changes to their internal caches though no details on exactly what those size changes encompassed were given to us. So, just know that a little tweaking did go on there. They have also done some tweaking to the framebuffer blending in their ROPs, so in some situations framebuffer blending could be faster.

Here is a chart that compares the number of transistors in each GPU.

Article Image

The G71 / 7900 GTX is actually still at the same power draw as the original G70 / 7800 GTX 256 GPU. When you look at the huge G71 frequency increase noted below, the equal power draw will make a bit more sense.

The reduction in transistors and the move to 90 nanometer process have allowed NVIDIA to crank up the frequency on these new GPUs. Here is a complete chart of what is new and different with these GPUs.

Article Image

The frequency on the GeForce 7900 GTX is now at 650 MHz core speed and 1.6 GHz memory speed. The default amount of memory on the 7900 GTX is 512MB of fast GDDR3. This should provide a noticeable boost in performance even compared to the GeForce 7800 GTX 512MB. The suggested retail price for this card will come in between $499-$649.

The GeForce 7900 GT is the most enticing out the bunch, clocked at 450 MHz core speed and 1.32 GHz memory speed with 24 pixels per clock it is actually slightly faster than a GeForce 7800 GTX 256MB video card. The suggested retail price for this card will be between $249-$399.

In the mid-range NVIDIA is releasing the GeForce 7600 GT which is clocked at 560 MHz core speed and 1.4 GHz memory speed. With 5 vertex units, and 12 pixels per clock and 256MB of framebuffer RAM it will of course be slower but still very competitive specs for the price point it is targeting. The suggested retail price will be between $179-$229. The $179 price representing the possibility for 128MB versions.

There is also support now for “Quad SLI.” We will talk more about that later since there aren’t any add-in-board options available for this. Basically there will be a version of the GeForce 7900 series in Quad SLI. It will have different clock speeds and will be an OEM option at first. There will be more on this in a month or two.